Instructions to use callensxavier/v10g2-dqn-compiler-optimization with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- stable-baselines3
How to use callensxavier/v10g2-dqn-compiler-optimization with stable-baselines3:
from huggingface_sb3 import load_from_hub checkpoint = load_from_hub( repo_id="callensxavier/v10g2-dqn-compiler-optimization", filename="{MODEL FILENAME}.zip", ) - Notebooks
- Google Colab
- Kaggle
v10g2 — Correctness-Priced DQN Compiler-Flag Predictor (the positive result)
A Deep Q-Network that selects the best gcc flag (-O0/-O1/-O2/-O3/-Ofast/-Os) per C/C++
translation unit for execution + CPU time, with a hard safety gate (no numeric/functional
regression vs the -O0 reference). Best of 5 seeds (seed 1337, 2M one-step bandit episodes).
Headline — safe AND fast (5-seed, held-out, program-disjoint)
| Metric | Value |
|---|---|
| Utility / mean speedup vs -O2 | 1.174× ± 0.011 |
| Numeric regression | 0.0% (all 5 seeds, bootstrap 95% CI [0,0]) |
| Functional regression | 0.0% |
| Accuracy (best-flag) | 55.6% ± 0.7 |
This is the result the v10F preprint set out to obtain. v10F was an honest negative result — a
speed-weighted agent degenerated into an -Ofast-chaser (≈20% numeric regressions). v10g2 prices
correctness above speed (numeric_fail penalty 200 + an -Ofast-risk term) and trains on a
corpus deliberately balanced across the fast-math safety boundary, so the agent finds safe
speedups instead of chasing unsafe ones.
Reward-mode ablation (held-out, n=15) — why this works
| reward | utility | numeric reg | 95% CI |
|---|---|---|---|
| speed_only | 1.232× | 13.3% | [0, 33] |
| v10f (partial penalty) | 1.201× | 13.3% | [0, 33] |
| correctness_dominant | 1.189× | 0.0% | [0, 0] |
Pricing correctness above speed drives regressions to zero for ~4% of mean speedup; a partial penalty (v10f) is not enough.
Lineage
v10F speed-weighted (1.069×, ~20% reg, Ofast-chaser) → v10g correctness/small corpus (1.006×, 10%) → v10g2 correctness/balanced corpus (1.174×, 0%).
Files
v10g2_dqn_seed1337_BEST_2M.zip (SB3 DQN, best seed), dataset_v10g2_public.csv (PolyBench×2,
micro, fast-math-balanced C++, OpenMP — internal TUs excluded), split_public.json, scaler.json,
EVAL_SUMMARY.md, metadata.json. Papers: PAPER_V10H_PIVOT_AND_NOVEL.md (PPO→DQN pivot proofs +
novel CVaR-DQN), PAPER_V10F_ARXIV.md (honest methodology).
Safe use
Serve with the ofast_guarded floor (default): -Ofast only when verified safe for the input, else
-O3. The bare argmax is conservative here; the floor guarantees the served flag is never
-Ofast-unsafe.
- Downloads last month
- 431